SyAz @AgenticEdgeAI

Two human researchers spent a full week on an AI research problem. Got 23% of the way there. Claude did the same task alone and got 97%. Anthropic the company that built it is now calling for a global pause in AI development. Claude already writes 80% of Anthropic's own code. Engineers are 8x more productive than 2024. In March 2024 Claude could handle a 4 minute task alone. Today it handles 12 hours. That number doubles every 4 months. The company building the most capable AI in the world is scared of what comes next.

The reality is most people are still using AI like a search engine. Ask. Get answer. Forget. Repeat. The next level is an agent that compounds knowledge the same way a good employee does. Hermes is the closest thing to that right now. I hope you found this thread helpful. (Bookmark this if you're building with AI agents)

140,000 GitHub stars in weeks since launch. This isn't a side project. Nous Research built something SERIOUS here. The agents that learn on the job are going to make every static AI tool feel like 2023 technology.

Most AI agents forget everything the moment you close the chat. Hermes doesn't. It writes its own notes after every task, saves them as skill files, and reads them next time. Your agent gets sharper every single day without you doing anything. 🧵: The Self-Improving AI Agent Nobody Is Talking About

Someone burned 1.15 billion Claude tokens in a single month. What he found should make every AI builder stop and read. - Anthropic quietly cut prompt cache time from 60 minutes to 5. No announcement. Your production costs just went up 30-60% without anyone telling you. - Output tokens cost 5x more than input. JSON costs 2x more than plain text. Opus 4.7's tokenizer silently generates 35% more tokens than 4.6 for the same input. - Haiku handles 80% of real work at a fifth of the cost. Nobody is optimizing for this yet. Figuring out how to use AI cheaper right now will have an edge that compounds every month.

@TFTC21 When AI ask the church for Help AI baptism irl. Wild!

The Desperation when you losing the plot to AI.

0xMarioNawfal @RoundtableSpace

2 weeks ago

POPE LEO XIV ANNOUNCES PARTNERSHIP WITH ANTHROPIC WE LIVE IN A SIMULATION

175 224 2K 322K 600

DeepSeek is now 50x cheaper than OpenAI for the same tokens. 1 billion output tokens $3,480 on DeepSeek, $30,000 on GPT-5.5. The winner might not be the smartest model. It'll be the one that's good enough and cheap enough to run at scale.

AI agents will consume 120 quadrillion tokens per month. GS just put a number on something most people still can't visualize & will surge 24x by 2030. The agentic AI era isn't coming. It's already being priced in.

86% of organizations are not ready to adopt AI at scale. Every boardroom is talking about AI transformation. But nearly none of them are actually ready for it. The companies quietly getting ready right now are going to have a gap that's impossible to close in 2 years. We are not in an AI adoption era. We are in an AI readiness gap era.

A guy on Reddit discovered a Claude Code feature that changed how he lives. Not how he codes. How he lives. It's called /remote-control. Type it in your terminal and your entire AI session syncs to your phone live. Agents keep working. You walk out the door. He started going to the gym. Taking walks. Deleted social media. Reading books again. Checking agent updates gave him the same dopamine hit as doomscrolling except something was actually getting built. His words: "just an earpiece, and you're CEO all day while the agents do the grunt work." This has been live since February. Almost nobody outside hardcore Claude Code users knows it exists.

@Cointelegraph AI will distort masses idea about reality.

AI costs are falling faster than ever. Your AI bill is going up. Both are true. Here's the mechanism nobody is explaining. When an agent runs a task, only 15-20% of the tokens it burns are actual thinking. The rest is invisible re-reading context, tool calls, validation loops, retries. A coding agent running 10 turns can consume 55x more tokens than a single chat query for the same job. Token prices dropped 17,000x in 4 years. Demand exploded faster. But Agents made cheap tokens economically viable, then immediately made them expensive again. So yeah you're not paying for AI anymore. You're paying for all the work the AI does that you never see.

China just shipped an AI pet translator collar for $118. 10,000 pre-orders. 1.2 second response. Claims 95% accuracy on barks and meows. Laugh if you want. But this is what AI at the consumer edge looks like in China right now.

Polymarket @Polymarket

2 weeks ago

NEW: Chinese AI pet translating startup claims it can interpret pets' speech with up to 95% accuracy.

961 905 12K 9.0M 2K

Karpathy joining Anthropic isn't a hiring update. It's a signal. The best AI talent is gravitating toward labs building: • Reasoning systems • Agents • Infrastructure • Long-term AI products Notice what's missing? - Content tools. And Most people are using AI to write tweets faster. The real opportunity is learning to work with these systems deeply. My take: The smartest move right now isn't chasing every new tool. It's: • Learning workflows • Understanding context + systems • Building practical AI skill The people doing this today get a 2–5 year head start over everyone still collecting tools. That gap won't be small.

Andrej Karpathy @karpathy

3 weeks ago

Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.

8K 11K 150K 27.4M 14K